移動向量精煉新方法運用於視訊處理和視訊轉換編碼的研究

Abstract

在這篇論文中，我們提出了修正視訊串流中移動向量特徵的新方法，並且實作幾個以移動向量為基礎的應用。我們將討論與驗證在MPEG視訊串流中的物件偵測與擷取，以物件為基礎的視訊串流，以及在有限資源環境中的視訊轉換編碼。我們對直接由壓縮域中的視訊串流擷取出的位移向量做處理，以降低移動向量的雜訊並獲得更多的物件資訊。透過我們所提出的系統，包含空間資訊過濾元件、時間資訊過濾元件以及紋路過濾元件，可以使這些物件資訊變得更精確。此外我們針對特定的應用提出了適應性移動向量修正的方法。因此，我們提出的視訊處理與通訊演算法可以準確地取得使用者想要的結果，在執行時間上也更有效率。我們也針對系統效能與其他常用的相關研究及技術作比較。 我們針對MPEG7測試資料與額外的視訊測試序列(超過200小時)來進行實驗，並用標準的指標評估系統效能，實驗結果顯示我們提出的系統效能明顯的優於其他技術。我們也介紹了新的效能評估指標來正確地追蹤我們的系統效率。 除此之外，本篇論文中也描述了我們發展的使用者介面，讓使用者可以維護與監控演算法的執行過程。

In this thesis we propose novel approaches for refining motion vector features in video streams. Several motion vector based applications have been proposed and implemented. Object detection and extraction in MPEG video streams, object based video steaming, and video transcoding with resource limited environment are all discussed, designated and verified. Rather than processing the extracted motion vector fields directly extracted from video streams in the compressed domain, we perform several operations over the extracted motion vectors, in order to reduce the noise within the motion vector content and to obtain more robust object information. The information is refined through our proposed system which is composed of a spatial filter component, temporal filter component and texture filter component. In addition, an adaptive motion vector refinement has been proposed for specific applications. As a result, our proposed video processing schemes are more capable of accurately gaining desired results with more efficient performance in terms of runtime. We compare the performance of our proposed system with other popular and commonly related work and techniques. Based on the experimental results performed over the MPEG7 testing dataset and additional video testing sequences –more than 200 hours of videos-, the performance measured by standard performance metrics using our proposed system is superior to the alternative techniques. Moreover, we introduce new performance metrics to trace the efficiency of the proposed systems accurately. A user system interfaces is also presented, where users can maintain and monitor the process of the proposed algorithms.